Skip to main content
SECURITY PLATFORM // 02Actively Developed / Portfolio Project

Cyber Sentinel

Secure Real-Time Communication Platform

Cyber Sentinel is a resilient, real-time messaging architecture combining Spring Boot, Redis Pub/Sub, and STOMP WebSockets with proactive threat detection. By pairing high-speed K-Means anomaly clustering with Google Gemini AI contextual analysis, the platform intercepts malicious API floods, credential brute-forcing, and anomalous traffic patterns at the gateway before they degrade messaging infrastructure.

// REAL-TIME CORE

Spring Boot & STOMP WebSockets

Full-duplex bidirectional communication with Redis Pub/Sub broadcast and TTL-based ephemeral room reclamation.

// STATISTICAL ML PRE-FILTER

K-Means Anomaly Clustering

High-speed Euclidean outlier distance calculation flagging volumetric traffic surges and protocol anomalies in microseconds.

// LLM THREAT REASONING

Google Gemini AI Analysis

Synthesizes packet signatures, user agent distributions, and request entropy to execute automated edge mitigations.

// DUAL-LANE TOPOLOGY

Messaging & Threat Mitigation Architecture

Parallel inspection isolating legitimate WebSocket/STOMP traffic from edge threats before backend message propagation.

DUAL-LANE SECURITY & MESSAGING ARCHITECTURE // 2D SCHEMATICTHREAT STATE: MONITORING
LANE 1: AUTHORIZED REAL-TIME MESSAGING (STOMP / REDIS PUB-SUB)
STAGE 01
Authenticated Clients

Browser & mobile clients connecting over secure WSS

STAGE 02
WebSocket / STOMP

Full-duplex frame negotiation with heartbeat monitoring

STAGE 03
Spring Cloud Gateway

Centralized TLS termination, auth validation, and routing

STAGE 04
Redis Pub/Sub + TTL

Distributed room broadcasting with automatic TTL expiry

STAGE 05
Active Chat Engine

Message delivery to subscribed participants

LANE 2: ANOMALY DETECTION & AUTOMATED MITIGATION PIPELINE
01. INGRESS
API Rate Flood

Rate Violation

02. K-MEANS
Statistical Clustering

Outlier Cluster C4

03. GEMINI AI
Context Reasoning

Pattern Identified

04. MITIGATION
Connection Severed

Edge Filter Applied

// ATTACK SIMULATION HUD

Simulated Threat Ingress & Edge Interception

Trigger high-frequency API floods, brute-force dictionary attacks, volumetric spikes, or protocol fuzzing to inspect automated mitigation.

SIMULATE REAL-TIME THREAT ATTACK // LIVE GATEWAY MITIGATION
Select a threat signature to trigger ML clustering & AI interception
// K-MEANS STATISTICAL CLUSTERINGWEKA ML PIPELINE

Unsupervised Outlier Centroid Isolation

Traffic vectors projected across request frequency, entropy variance, and socket persistence metrics.

Cluster C1 (Normal)
Cluster C2
Euclidean Distance:4.82 Euclidean distance (Threshold: 2.10)
Signature:850 req/sec from single CIDR block | WebSocket frame payload identical
Request Frequency: ExtremeEntropy: 0.04 (Repetitive)Socket Age: 1.2s
// GOOGLE GEMINI AI REASONING PASSCONTEXTUAL EVALUATION

Automated Threat Synthesis & Intent Classification

Volumetric packet flooding originating from automated bot network masquerading as legitimate chat clients. Exceeds standard human typing entropy bounds by 99.4%.
Inference Latency: 142ms // Confidence: High
// AUTOMATED EDGE FILTER MITIGATION

Spring Cloud Gateway Action Executed

Dynamic IP throttle applied at Spring Cloud Gateway; socket connection dropped and ephemeral room key invalidated.

Socket Status: TERMINATEDTraffic Normalized
TECHNOLOGY STACK// RESILIENT REAL-TIME SYSTEM

Categorized Architecture Stack

// BACKEND SERVICES
JavaSpring BootSpring Cloud Gateway
// REAL-TIME & CACHE
WebSocketSTOMP ProtocolRedis (Pub/Sub + TTL)
// MACHINE LEARNING & AI
Weka (K-Means Clustering)Markov Chain ModelingGoogle Gemini AI
// FRONTEND & UI
ReactTypeScriptTailwind CSSFramer Motion
// DEVOPS & INFRASTRUCTURE
DockerDocker Compose
ENGINEERING RATIONALE// ARCHITECTURAL TRADE-OFFS

Key Systems Decisions & Justifications

Why Redis Pub/Sub + TTL for chat rooms?
Decision: In-memory Pub/Sub with automatic key expiration

Temporary chat rooms do not need permanent relational database storage. Redis Pub/Sub delivers sub-millisecond multi-client broadcast while TTL keys automatically clean up dormant rooms without requiring expensive periodic database cron purges.

Production Rationale Verified
Why a centralized Spring Cloud Gateway?
Decision: Edge routing and unified rate inspection

Terminating WebSocket handshakes and client traffic at a dedicated gateway layer prevents unauthenticated packets from ever reaching backend services. It centralizes IP rate-limiting, authentication, and threat inspection in a single perimeter.

Production Rationale Verified
Why pair K-Means clustering with Google Gemini AI?
Decision: Dual-layer detection: statistical ML pre-filter + LLM contextual intelligence

Sending raw traffic directly to an LLM is cost-prohibitive and slow. Unsupervised K-Means clustering flags statistical outliers in microsecond timeframes. Once an outlier is flagged, Gemini AI analyzes the packet narrative and user context to distinguish malicious attacks from benign flash crowds.

Production Rationale Verified